Do autonomous vehicle companies need to disclose AI decision logic to passengers?

Yes. Article 50(1) requires that natural persons interacting with an AI system are informed of that fact. Passengers must know the vehicle operates autonomously, its current autonomy level, and its operational capabilities and limitations. They do not need to understand the technical architecture, but must receive meaningful information about what the AI can and cannot do.

What does Article 13 require for autonomous vehicle AI?

Instructions for use must include: the provider's identity, the system's operational design domain (weather, road types, speed ranges, geographic limits), performance metrics (detection accuracy, reaction times), known limitations and failure modes, degraded mode behaviour, human takeover procedures, and guidance for fleet operators on monitoring and intervention.

What is the compliance deadline for autonomous system transparency?

High-risk autonomous systems must comply by August 2, 2026. The Digital Omnibus proposal (March 2026) may delay high-risk obligations to December 2027, but this has not been formally adopted. Given the intersection with type approval and EASA certification timelines, autonomous system providers should plan for August 2026.

Does Article 13 apply to ADAS systems that are not fully autonomous?

Advanced driver assistance systems (ADAS) that function as safety components of vehicles fall under Annex III, Section 2(a) and are classified as high-risk. Article 13 transparency requirements apply to these systems, including disclosure of capabilities, limitations, and the expected driver role. The General Safety Regulation already requires some of this transparency, and Article 13 extends it to the AI system level.

Article 13 Transparency for Autonomous Systems — EU AI Act

How EU AI Act Article 13 transparency applies to autonomous systems — operational design domain disclosure, passenger notification, fleet operator instructions.

Published April 2, 2026

Article 13 of the EU AI Act requires high-risk AI systems to be sufficiently transparent for deployers to interpret outputs and use them appropriately. For autonomous systems — self-driving vehicles, drones, and autonomous machinery — transparency determines whether human operators can safely monitor, intervene in, and trust AI-driven decisions in safety-critical environments. Under Annex III, Section 2(a), AI systems intended as safety components in road traffic management and autonomous vehicle operation are classified as high-risk.

This page explains how Article 13 transparency requirements apply specifically to autonomous systems — from self-driving vehicles to industrial drones — including operational domain disclosure, fleet operator instructions, passenger notification, and what code-level violations Scanara detects in autonomous system software.

Which Autonomous AI Systems Must Be Transparent?

Self-Driving Vehicles (SAE L3-L5)

Autonomous vehicles must disclose their operational design domain (ODD) boundaries to passengers and fleet operators. Passengers must know the system's capabilities and limitations — including when the AI may request human takeover and the expected transition time.

Autonomous Drones and UAVs

AI-assisted detect-and-avoid (DAA) systems and autonomous flight planning must provide operators with transparent explanations of navigation decisions, particularly when the AI overrides planned routes or triggers emergency procedures. EASA requires transparency about AI-assisted DAA limitations.

Autonomous Industrial Machinery

AI-controlled warehouse robots, agricultural machinery, and manufacturing systems must provide operators with clear information about the system's decision logic, safety boundaries, and conditions under which the AI may behave unpredictably or require human intervention.

Autonomous Maritime and Rail Systems

AI navigation systems for autonomous ships and rail operations must be transparent about routing decisions, collision avoidance logic, and operational limits. Transport authorities must receive sufficient information to evaluate the AI's fitness for deployment in specific operational contexts.

Key Transparency Obligations for Autonomous Systems

Article 13 requires instructions for use that enable deployers to interpret AI outputs appropriately. For autonomous systems, this means fleet operators and transport authorities must understand the AI's operational boundaries, decision rationale, and behaviour under edge cases — transparency that directly affects physical safety.

Operational Design Domain Must Be Disclosed

Instructions for use must clearly define the ODD — the specific conditions under which the autonomous system is designed to operate safely. This includes weather conditions, road types, speed ranges, geographic boundaries, and time-of-day restrictions. Operating outside the documented ODD without disclosure violates Article 13.

Deployers Must Understand Intervention Points

Article 13(1) requires deployers to interpret outputs appropriately. For autonomous systems, this means fleet operators must understand when and how to intervene — including takeover request timings, fallback behaviours, and the system's expected response when operating at the edge of its capabilities.

Passengers Must Know System Capabilities

Passengers in autonomous vehicles must be informed of the system's operational capabilities and limitations. This includes the current autonomy level, what the AI can and cannot handle, and what is expected of the human driver (if any). Vehicle identification markings may vary by jurisdiction.

Edge Case Behaviour Must Be Documented

Technical documentation must explain system behaviour under degraded conditions: sensor failures, adverse weather, unusual obstacles, and network connectivity loss. Article 13(3)(b)(i) requires disclosure of limitations and conditions where the system may malfunction — for autonomous systems, these are safety-critical disclosures.

Regulatory Overlaps

Autonomous system transparency intersects with type approval frameworks, aviation regulations, and safety standards. Article 13 must be coordinated with existing sector-specific transparency and documentation requirements.

Type Approval Framework Regulation (TAFR)

The EU Type Approval Framework Regulation requires detailed technical documentation for autonomous vehicle certification, including the operational design domain, safety validation data, and human-machine interface design. AI Act Article 13 requirements for instructions for use must be aligned with TAFR documentation to avoid conflicting or duplicative disclosures.

General Safety Regulation (GSR)

The GSR mandates advanced driver assistance systems and sets requirements for their transparency to drivers. Event data recorders, driver drowsiness detection, and intelligent speed assistance all require driver notification. Article 13 extends these transparency requirements to the AI system level, requiring documentation of capabilities, limitations, and performance.

EASA Air Operations Regulations

The European Union Aviation Safety Agency regulates AI in aviation including autonomous drone operations. EASA requires transparency about AI-assisted detect-and-avoid decisions and their limitations. Article 13 adds requirements for system-level documentation of AI capabilities, accuracy, and robustness that supplement EASA's operational transparency framework.

ISO 21448 (SOTIF) — Safety of the Intended Functionality

ISO 21448 addresses safety risks from functional insufficiencies in autonomous systems, requiring identification and documentation of triggering conditions and failure modes. Article 13 transparency requirements for disclosing system limitations align closely with SOTIF's documentation of known and unknown unsafe scenarios.

Common Violations Scanara Detects

Missing ODD Boundary Documentation in System Output

Autonomous navigation code that makes routing or control decisions without referencing or exposing the operational design domain boundaries to the operator interface. Article 13 requires deployers to understand system limits — Scanara flags decision modules that lack ODD parameter exposure.

No AI Interaction Disclosure in Passenger Interface

Vehicle or drone passenger interfaces that display navigation or operational status without disclosing that decisions are AI-driven. Article 50(1) requires notification when natural persons interact with AI systems — passengers must know the system's autonomy level and current capabilities.

Instructions for Use Missing Degraded Mode Behaviour

System documentation that describes normal operation without documenting behaviour under sensor failure, adverse weather, connectivity loss, or other degraded conditions. Article 13(3)(b)(i) requires disclosure of limitations and conditions where the system may malfunction — critical for autonomous system safety.

Article 13 Compliance Checklist for Autonomous Systems

Art. 13(1)

Make AI Decisions Interpretable for Operators

Ensure fleet operators and transport authorities can understand the AI's decision rationale, including navigation choices, speed adjustments, and obstacle avoidance decisions. Provide real-time and post-hoc interpretability tools.

Art. 13(3)(b)

Document Operational Design Domain

Provide detailed documentation of the ODD including weather, road type, speed range, geographic, and temporal boundaries. Include clear guidance on what happens when the system approaches or exceeds ODD limits.

Art. 13(3)(b)(i)

Disclose Degraded Mode and Edge Case Behaviour

Document system behaviour under sensor failure, adverse weather, connectivity loss, and unusual obstacle scenarios. Include fallback strategies, minimum risk conditions, and expected transition times for human takeover requests.

Art. 50(1)

Notify Passengers and Other Road Users

Inform passengers of the system's autonomy level, current capabilities, and what is expected of them. For other road users, implement vehicle identification markings as required by jurisdiction.

Art. 13(3)(b)(ii)

Disclose Safety and Performance Metrics

Report system performance including object detection accuracy, false positive/negative rates, reaction times, and safety validation results across the documented ODD. Include performance under edge conditions.

Frequently Asked Questions

How Scanara Helps

Scanara automates EU AI Act compliance from code to dossier. Connect your GitHub repos and get compliance reports in minutes.